How the KernelX Score works
The full methodology behind the score — published, versioned, and deterministic, so every number can be traced to evidence.
What it measures
Five pillars, one hundred points
The KernelX Score (0–100) measures how well a brand's observable surface supports being correctly understood by AI systems. It is computed from deterministic evidence collected from your homepage, metadata, structured data, machine-access files and linked profiles — the same raw material AI crawlers and assistants actually see.
- Identity & structured data — 30 points. Organization JSON-LD completeness, title quality, share cards, canonical, name consistency, favicon.
- Copy understandability — 25 points. Whether the hero states what you are and who you serve; H1 clarity; descriptive navigation; discoverable contact.
- Machine access — 20 points. HTTPS, a permissive robots.txt, sitemap.xml, llms.txt, response health.
- Corroboration — 15 points. Linked social profiles, structured sameAs links, legal pages.
- Entity Clarity™ — 10 points. Cross-surface identity consistency, scaled into the composite.
Every factor reports the points it earned, the points possible, and the evidence it examined. A score is never asserted without its arithmetic.
Entity Clarity™
Do all your surfaces describe the same company?
Entity Clarity measures how consistently AI systems would infer the same identity, category and purpose across your surfaces: title, meta description, Open Graph, Organization structured data, hero copy and footer. We extract identity language from each surface independently, compute pairwise agreement, and weight the pairs that bridge structured data and visible copy most heavily — because that divergence is precisely what machines experience. The result is a 0–100 score with the strongest and weakest surface pairs quoted in the report.
Confidence
Confidence is derived, never asserted
Each report's confidence level (High / Medium / Low) is computed from evidence coverage: how many surfaces could be fetched and how many expected signals resolved, alongside structured-data quality and external corroboration. When we could observe less, the report says so and claims less. The derivation ships inside every report as data.
What it doesn't measure
Honest limitations
- Actual AI behaviour. The score measures your observable surface, not live model outputs. Measuring what assistants actually say requires repeated sampling across model families — that is the KernelX platform's job, with its own published methodology.
- JavaScript-rendered content. We analyze served HTML, as most crawlers do. If your content only exists after JavaScript runs, the report will honestly reflect the thinner surface — which is itself a finding.
- Off-site authority. Backlinks, press and community reputation are out of scope in v1.0.
- Synonym awareness. Entity Clarity v1.0 measures lexical agreement; two phrasings of one idea can read as divergence.
Version history
Versions
Methodology v1.0 (July 2026) — initial public release: five-pillar rubric, Entity Clarity v1.0.0, derived confidence, deterministic engine v1.0.0. Every report carries the full version fingerprint of the engine, methodology, weights and Entity Clarity that produced it; any change to weights requires a version bump and a re-baselined benchmark corpus.